Q-Learning Gridworlds

A reinforcement learning assignment built on Gymnasium. The harness around the learning is all there: environment setup for Taxi and two Frozen Lake variants, an epsilon-greedy training loop, a separate evaluation loop that measures average discounted reward per episode, policy printing for debugging, and Plotly charting of the results. The piece that is missing is the Q-table itself.

Attention: Archived as unfinished. The QValues class is still the assignment skeleton: the update rule, the lookup, and the best-action selection are all TODO, so the agent currently learns nothing.